fix for generation aware retries issue

operator-framework/src/main/java/com/github/containersolutions/operator/processing/EventScheduler.java

@@ -64,19 +64,22 @@ void scheduleEventFromApi(CustomResourceEvent event) {
         try {
             lock.lock();
             log.debug("Scheduling event from Api: {}", event);
-            if (event.getAction() == Action.DELETED) {
-                // This removes data from memory for deleted resource (prevent memory leak basically).
-                // Its quite interesting that this is always sufficient here (no finalizer or other mechanism needs to be involved).
-                // Thus, if operator is running we get DELETE the event, if not the memory is already gone anyways.
-                eventStore.removeLastGenerationForDeletedResource(event.resourceUid());
-                if (event.getResource().getMetadata().getDeletionTimestamp() != null) {
-                    // Note that we always use finalizers, we want to process delete event just in corner case,
-                    // when we are not able to add finalizer (lets say because of optimistic locking error, and the resource was deleted instantly).
-                    // We want to skip in case of finalizer was there since we don't want to execute delete method always at least 2x,
-                    // which would be the result if we don't skip here. (there is no deletion timestamp if resource deleted without finalizer.)
-                    log.debug("Skipping delete event since deletion timestamp is present on resource, so finalizer was in place.");
-                    return;
-                }
+            if (event.getAction() == Action.DELETED && event.getResource().getMetadata().getDeletionTimestamp() != null) {
+                // This removes data from memory for deleted resource (prevent memory leak).
+                // There is am extreme corner case when there is no finalizer, we ignore this situation now.
+                eventStore.cleanup(event.resourceUid());
+                // Note that we always use finalizers, we want to process delete event just in corner case,
+                // when we are not able to add finalizer (lets say because of optimistic locking error, and the resource was deleted instantly).
+                // We want to skip in case of finalizer was there since we don't want to execute delete method always at least 2x,
+                // which would be the result if we don't skip here. (there is no deletion timestamp if resource deleted without finalizer.)
+                log.debug("Skipping delete event since deletion timestamp is present on resource, so finalizer was in place.");
+                return;
+            }
+            if (generationAware) {
+                // we have to store the last event for generation aware retries, since if we received new events since
+                // the execution, which did not have increased generation we will fail automatically on a conflict
+                // on a retry.
+                eventStore.addLastEventForGenerationAwareRetry(event);
             }
             // In case of generation aware processing, we want to replace this even if generation not increased,
             // to have the most recent copy of the event.
@@ -145,13 +148,28 @@ void eventProcessingFailed(CustomResourceEvent event) {
                 scheduleNotYetScheduledEventForExecution(event.resourceUid());
             } else {
                 log.debug("Event processing failed. Attempting to re-schedule the event: {}", event);
-                scheduleEventForExecution(event);
+                if (generationAware) {
+                    CustomResourceEvent eventToRetry = selectEventToRetry(event);
+                    scheduleEventForExecution(eventToRetry);
+                } else {
+                    scheduleEventForExecution(event);
+                }
             }
         } finally {
             lock.unlock();
         }
     }
 
+    private CustomResourceEvent selectEventToRetry(CustomResourceEvent event) {
+        CustomResourceEvent lastEvent = eventStore.getReceivedLastEventForGenerationAwareRetry(event.resourceUid());
+        if (!event.getResource().getMetadata().getResourceVersion()
+                .equals(lastEvent.getResource().getMetadata().getResourceVersion())) {
+            return lastEvent;
+        } else {
+            return event;
+        }
+    }
+
     private void scheduleNotYetScheduledEventForExecution(String uuid) {
         CustomResourceEvent notScheduledEvent = eventStore.removeEventNotScheduled(uuid);
         scheduleEventForExecution(notScheduledEvent);

operator-framework/src/main/java/com/github/containersolutions/operator/processing/EventStore.java

@@ -8,6 +8,7 @@ public class EventStore {
     private final Map<String, CustomResourceEvent> eventsNotScheduled = new HashMap<>();
     private final Map<String, CustomResourceEvent> eventsUnderProcessing = new HashMap<>();
     private final Map<String, Long> lastGeneration = new HashMap<>();
+    private final Map<String, CustomResourceEvent> receivedLastEventForGenerationAwareRetry = new HashMap<>();
 
     public boolean containsNotScheduledEvent(String uuid) {
         return eventsNotScheduled.containsKey(uuid);
@@ -52,7 +53,16 @@ public Long getLastStoredGeneration(CustomResourceEvent event) {
         return lastGeneration.get(event.getResource().getMetadata().getUid());
     }
 
-    public void removeLastGenerationForDeletedResource(String uuid) {
+    public void addLastEventForGenerationAwareRetry(CustomResourceEvent event) {
+        receivedLastEventForGenerationAwareRetry.put(event.resourceUid(), event);
+    }
+
+    public CustomResourceEvent getReceivedLastEventForGenerationAwareRetry(String uuid) {
+        return receivedLastEventForGenerationAwareRetry.get(uuid);
+    }
+
+    public void cleanup(String uuid) {
         lastGeneration.remove(uuid);
+        receivedLastEventForGenerationAwareRetry.remove(uuid);
     }
 }

operator-framework/src/test/java/com/github/containersolutions/operator/ControllerExecutionIT.java

@@ -9,6 +9,7 @@
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
+import java.util.HashMap;
 import java.util.concurrent.TimeUnit;
 
 import static com.github.containersolutions.operator.IntegrationTestSupport.TEST_NAMESPACE;
@@ -36,6 +37,7 @@ public void configMapGetsCreatedForTestCustomResource() {
 
             awaitResourcesCreatedOrUpdated();
             awaitStatusUpdated();
+            assertThat(integrationTestSupport.numberOfControllerExecutions()).isEqualTo(2);
         });
     }
 
@@ -52,6 +54,26 @@ public void eventIsSkippedChangedOnMetadataOnlyUpdate() {
         });
     }
 
+    // We test the scenario when we receive 2 events, while the generation is not increased by the other.
+    // This will cause a conflict, and on retry the new version of the resource needs to be scheduled
+    // to avoid repeating conflicts
+    @Test
+    public void generationAwareRetryConflict() {
+        initAndCleanup(true);
+        integrationTestSupport.teardownIfSuccess(() -> {
+            TestCustomResource resource = testCustomResource();
+            TestCustomResource resource2 = testCustomResource();
+            resource2.getMetadata().getAnnotations().put("testannotation", "val");
+
+            integrationTestSupport.getCrOperations().inNamespace(TEST_NAMESPACE).create(resource);
+            integrationTestSupport.getCrOperations().inNamespace(TEST_NAMESPACE).createOrReplace(resource2);
+
+            awaitResourcesCreatedOrUpdated();
+            awaitStatusUpdated(10);
+        });
+    }
+
+
     void awaitResourcesCreatedOrUpdated() {
         await("configmap created").atMost(5, TimeUnit.SECONDS)
                 .untilAsserted(() -> {
@@ -63,7 +85,11 @@ void awaitResourcesCreatedOrUpdated() {
     }
 
     void awaitStatusUpdated() {
-        await("cr status updated").atMost(5, TimeUnit.SECONDS)
+        awaitStatusUpdated(5);
+    }
+
+    void awaitStatusUpdated(int timeout) {
+        await("cr status updated").atMost(timeout, TimeUnit.SECONDS)
                 .untilAsserted(() -> {
                     TestCustomResource cr = integrationTestSupport.getCrOperations().inNamespace(TEST_NAMESPACE).withName("test-custom-resource").get();
                     assertThat(cr).isNotNull();
@@ -78,6 +104,7 @@ private TestCustomResource testCustomResource() {
                 .withName("test-custom-resource")
                 .withNamespace(TEST_NAMESPACE)
                 .build());
+        resource.getMetadata().setAnnotations(new HashMap<>());
         resource.setKind("CustomService");
         resource.setSpec(new TestCustomResourceSpec());
         resource.getSpec().setConfigMapName("test-config-map");